jmo [he/him, comrade/them]

Taking the day off and protesting with my local socialist org.

Curious to see if/how this relates to the credit swap the US has been discussing with the UAE. If you can count on the Trump administration for anything it’s to be extremely transactional.

This has big, “the other leaders are from Canada, you wouldn’t know them” energy.

Ok, the numbers and spend question led me down a really interesting rabbit hole and I wanted to share a little bit of what I learned. TL;DR AI data center spend, in absolute and relative terms, is really high. Please take all my statements with a heavy dose of salt as I’m not an economist or professional researcher.

It’s probably less than railroads ~6% of GDP and probably more than telco spend leading up to the dotcom bubble ~1.5% of GDP. It’s reasonable to guess that is represents, or will by the end of 2026, ~ 2.2% of US GDP. Like the dotcom bubble it’s largely funded by private capital. Unlike the dotcom bubble it’s much more reliant on debt, specifically private credit debt, we’ll get back to why that might matter later. Where things get interesting is in the durable nature of the utility value of the goods produced by the investment. Railroad capital mostly has a 100+ year utility value, tracks can be used for a long time with some regular maintenance. Fiber has a 15+ year, probably a lot longer, utility value and again a lot of the associated infrastructure stays useful for a long time with maintenance. Data center infrastructure and GPUs are not so lucky. The people spending the money say the GPUs are good for 5 years, they probably have a 1-3 year usable lifespan. Data center infrastructure itself has a much longer useful lifespan but those buildings aren’t exactly general purpose, if AI is a bust you’ll need to have a new use for all that space. The so what of all that is this is looking even less rational and useful than the dotcom era infrastructure investment boom.

On the debt financing thing I talked about getting back to I’d suggest you take a look at this article that has the heterodox economist Michal Hudson talking about why he thinks the west is in for a really bad ride if AI + private credit + energy crisis all come together for a lemon party type situation.

Yes! I was just having this discussion last night. We have another 2.5 trillion in capex, the money corporations allocate to investing in their infrastructure, committed to AI datacenter buildout this year. That’s A LOT of money. It’s not being spent on bridges, roads, hospitals, schools, etc. It’s being spent on specialized data centers intended to house massive GPU farms to run and train the latest and greatest AI models.

AFAICT this only pays off they manage to actually beat the theoretically limitations of these LLMs and make a machine god. The so what of all that is we have A LOT of capital chasing, what is probably, a fantasy. As it does so it isn’t creating a useful stay behind asset. AI datacenters aren’t general purpose and can’t be easily reused. Add to that the fact that the GPUs don’t actually have a very long productive life and you start to see problems compound.

One important methodological note here: Zitron’s numbers on Capex are WAY better than mine. I did some digging into that 2.5 trillion number while writing this comment and found out that was based on a single source, Gartner, which has an incentive to pump up the numbers.

I read him and largely agree, I also work in the space. The thing I’d say we need to be careful of with his analysis is he consistent assumes the worst case scenario when there is ambiguity and he assumes that the worst possible outcomes from every negative event experienced by AI companies.

That’s not to say he isn’t useful, just be aware that he can easily go beyond what the evidence realistically supports and that can leave you overly optimistic about the downfall of these AI compainies.

Annualized ARR, So as an example Anthropic probably just had a great month where they made an extra 1-2 billion in revenue. They then go in multiply it by 12 and say that their annualized ARR increased by 12-24 billion. It’s misleading when subscription companies do it but at least for them users then to sign up to spend the same amount every month or year. With a consumption based model it’s even wonkier because if you had costs get out of control in April you’re likely to to try and bring them down by lowering consumption in May.

So I’ve just started reading MoA and would love to better understand what you’re saying. Who is Michael Tracey and why does he matter? I saw they had a post that implied Yemen had closed the Bab El-Mandeb which wasn’t correct but that was the biggest miss I’ve seen lately.

So I don’t get the US angle here. Why ask for a ceasefire when time only benefits Iran? Were Iranian attacks on US forces so effective that a 2 week pause allows them to material improve the balance of forces in the region? Do we have evidence of anywhere near enough buildup to launch an invasion? I don’t think the situation as far as interceptors and standoff munitions is sufficiently elastic for 2 weeks to make much difference.

I believe what your saying but it seems like everyday the straight isn’t open hurts the US side way more than the Iranian. Can anyone help me figure out what I’m missing?

So the real question will be does Iran let ships through without a toll? That money is key to rebuilding and throttling traffic is their best leverage.

It’s the best part about Trump more broadly. He strips away the veneer of empire in a way that’s hard to ignore.

So this is really interesting because I was just talking to an organization that is looking to build the exact type of systems the article describes. My thinking is at least part of it is marketing for in field autonomous drone manufacturing systems being designed by the big defense contractors. Not disagreeing with the article, just thought that was an important perspective to keep in mind.

It’s really funny to see how the MIC is fundamentally hampering western military capabilities as their economic strategies incentivize building the big expensive weapons systems which are proving to be significantly less effective than building more distributed and cost effective systems.

I think this ties into the key responsibility of those of us based in the imperial core. Disruption and pressure on the system from inside gives Iran more breathing room and better negotiating terms. Do folks have good ideas for things we can be doing while staying within the bounds of the law? The Internationalist group I’m involved in has been looking at helping students organize anti recruitment efforts. I’d love to hear about things other folks are trying.

This may be another, “do nothing then win” scenario for China. The AI bubble depends on continuous infusion of capital for the next several years. That effectively means a lot of folks need to borrow a lot of money. Interest rates increasing, the cost of power increasing, the costs of CPUs, GPUs, and memory increasing, and a slowdown in delivery times for all those components all exacerbate their core problem. We may well be looking at an AI bubble burst to go along with a broader economic downturn due to the war.

China doesn’t need to blockade anyone. Continue to be a reliable and trustworthy partner while the US flounders and see what happens. The only major economy that doesn’t depend on AI succeeding is China.